src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_sparc64_hpc_ace_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  91     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  92     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  93     real             *vftab;
  94     _fjsp_v2r8       itab_tmp;
  95     _fjsp_v2r8       dummy_mask,cutoff_mask;
  96     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  97     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  98     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  99
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_vdw->data;
 118     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _fjsp_setzero_v2r8();
 146         fiy0             = _fjsp_setzero_v2r8();
 147         fiz0             = _fjsp_setzero_v2r8();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _fjsp_setzero_v2r8();
 155         vvdwsum          = _fjsp_setzero_v2r8();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 173             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 174             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 180
 181             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 185             vdwjidx0A        = 2*vdwtype[jnrA+0];
 186             vdwjidx0B        = 2*vdwtype[jnrB+0];
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 196             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 197                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 198
 199             /* Calculate table index by multiplying r with table scale and truncate to integer */
 200             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 201             itab_tmp         = _fjsp_dtox_v2r8(rt);
 202             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 203             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 204             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 205
 206             vfconv.i[0]     *= 8;
 207             vfconv.i[1]     *= 8;
 208
 209             /* COULOMB ELECTROSTATICS */
 210             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 211             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 212
 213             /* CUBIC SPLINE TABLE DISPERSION */
 214             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 215             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 216             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 217             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 218             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 219             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 220             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 221             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 222             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 223             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 224             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 225
 226             /* CUBIC SPLINE TABLE REPULSION */
 227             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 228             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 229             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 230             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 231             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 232             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 233             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 234             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 235             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 236             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 237             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 238             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 239             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 243             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 244
 245             fscal            = _fjsp_add_v2r8(felec,fvdw);
 246
 247             /* Update vectorial force */
 248             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 249             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 250             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 251
 252             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 253
 254             /* Inner loop uses 66 flops */
 255         }
 256
 257         if(jidx<j_index_end)
 258         {
 259
 260             jnrA             = jjnr[jidx];
 261             j_coord_offsetA  = DIM*jnrA;
 262
 263             /* load j atom coordinates */
 264             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 265                                               &jx0,&jy0,&jz0);
 266
 267             /* Calculate displacement vector */
 268             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 269             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 270             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 274
 275             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 276
 277             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 278
 279             /* Load parameters for j particles */
 280             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 281             vdwjidx0A        = 2*vdwtype[jnrA+0];
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 291             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 292                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 293
 294             /* Calculate table index by multiplying r with table scale and truncate to integer */
 295             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 296             itab_tmp         = _fjsp_dtox_v2r8(rt);
 297             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 298             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 299             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 300
 301             vfconv.i[0]     *= 8;
 302             vfconv.i[1]     *= 8;
 303
 304             /* COULOMB ELECTROSTATICS */
 305             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 306             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 307
 308             /* CUBIC SPLINE TABLE DISPERSION */
 309             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 310             F                = _fjsp_setzero_v2r8();
 311             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 312             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 313             H                = _fjsp_setzero_v2r8();
 314             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 315             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 316             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 317             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 318             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 319             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 320
 321             /* CUBIC SPLINE TABLE REPULSION */
 322             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 323             F                = _fjsp_setzero_v2r8();
 324             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 325             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 326             H                = _fjsp_setzero_v2r8();
 327             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 328             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 329             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 330             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 331             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 332             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 333             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 334             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 338             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 339             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 340             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 341
 342             fscal            = _fjsp_add_v2r8(felec,fvdw);
 343
 344             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 345
 346             /* Update vectorial force */
 347             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 348             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 349             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 350
 351             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 352
 353             /* Inner loop uses 66 flops */
 354         }
 355
 356         /* End of innermost loop */
 357
 358         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 359                                               f+i_coord_offset,fshift+i_shift_offset);
 360
 361         ggid                        = gid[iidx];
 362         /* Update potential energies */
 363         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 364         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 365
 366         /* Increment number of inner iterations */
 367         inneriter                  += j_index_end - j_index_start;
 368
 369         /* Outer loop uses 9 flops */
 370     }
 371
 372     /* Increment number of outer iterations */
 373     outeriter        += nri;
 374
 375     /* Update outer/inner flops */
 376
 377     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*66);
 378 }
 379 /*
 380  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 381  * Electrostatics interaction: Coulomb
 382  * VdW interaction:            CubicSplineTable
 383  * Geometry:                   Particle-Particle
 384  * Calculate force/pot:        Force
 385  */
 386 void
 387 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 388                     (t_nblist                    * gmx_restrict       nlist,
 389                      rvec                        * gmx_restrict          xx,
 390                      rvec                        * gmx_restrict          ff,
 391                      struct t_forcerec           * gmx_restrict          fr,
 392                      t_mdatoms                   * gmx_restrict     mdatoms,
 393                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 394                      t_nrnb                      * gmx_restrict        nrnb)
 395 {
 396     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 397      * just 0 for non-waters.
 398      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 399      * jnr indices corresponding to data put in the four positions in the SIMD register.
 400      */
 401     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 402     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 403     int              jnrA,jnrB;
 404     int              j_coord_offsetA,j_coord_offsetB;
 405     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 406     real             rcutoff_scalar;
 407     real             *shiftvec,*fshift,*x,*f;
 408     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 409     int              vdwioffset0;
 410     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 411     int              vdwjidx0A,vdwjidx0B;
 412     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 413     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 414     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 415     real             *charge;
 416     int              nvdwtype;
 417     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 418     int              *vdwtype;
 419     real             *vdwparam;
 420     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 421     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 422     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 423     real             *vftab;
 424     _fjsp_v2r8       itab_tmp;
 425     _fjsp_v2r8       dummy_mask,cutoff_mask;
 426     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 427     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 428     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 429
 430     x                = xx[0];
 431     f                = ff[0];
 432
 433     nri              = nlist->nri;
 434     iinr             = nlist->iinr;
 435     jindex           = nlist->jindex;
 436     jjnr             = nlist->jjnr;
 437     shiftidx         = nlist->shift;
 438     gid              = nlist->gid;
 439     shiftvec         = fr->shift_vec[0];
 440     fshift           = fr->fshift[0];
 441     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 442     charge           = mdatoms->chargeA;
 443     nvdwtype         = fr->ntype;
 444     vdwparam         = fr->nbfp;
 445     vdwtype          = mdatoms->typeA;
 446
 447     vftab            = kernel_data->table_vdw->data;
 448     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 449
 450     /* Avoid stupid compiler warnings */
 451     jnrA = jnrB = 0;
 452     j_coord_offsetA = 0;
 453     j_coord_offsetB = 0;
 454
 455     outeriter        = 0;
 456     inneriter        = 0;
 457
 458     /* Start outer loop over neighborlists */
 459     for(iidx=0; iidx<nri; iidx++)
 460     {
 461         /* Load shift vector for this list */
 462         i_shift_offset   = DIM*shiftidx[iidx];
 463
 464         /* Load limits for loop over neighbors */
 465         j_index_start    = jindex[iidx];
 466         j_index_end      = jindex[iidx+1];
 467
 468         /* Get outer coordinate index */
 469         inr              = iinr[iidx];
 470         i_coord_offset   = DIM*inr;
 471
 472         /* Load i particle coords and add shift vector */
 473         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 474
 475         fix0             = _fjsp_setzero_v2r8();
 476         fiy0             = _fjsp_setzero_v2r8();
 477         fiz0             = _fjsp_setzero_v2r8();
 478
 479         /* Load parameters for i particles */
 480         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 481         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 482
 483         /* Start inner kernel loop */
 484         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 485         {
 486
 487             /* Get j neighbor index, and coordinate index */
 488             jnrA             = jjnr[jidx];
 489             jnrB             = jjnr[jidx+1];
 490             j_coord_offsetA  = DIM*jnrA;
 491             j_coord_offsetB  = DIM*jnrB;
 492
 493             /* load j atom coordinates */
 494             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 495                                               &jx0,&jy0,&jz0);
 496
 497             /* Calculate displacement vector */
 498             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 499             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 500             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 501
 502             /* Calculate squared distance and things based on it */
 503             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 504
 505             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 506
 507             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 508
 509             /* Load parameters for j particles */
 510             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 511             vdwjidx0A        = 2*vdwtype[jnrA+0];
 512             vdwjidx0B        = 2*vdwtype[jnrB+0];
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 522             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 523                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 524
 525             /* Calculate table index by multiplying r with table scale and truncate to integer */
 526             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 527             itab_tmp         = _fjsp_dtox_v2r8(rt);
 528             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 529             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 530             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 531
 532             vfconv.i[0]     *= 8;
 533             vfconv.i[1]     *= 8;
 534
 535             /* COULOMB ELECTROSTATICS */
 536             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 537             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 538
 539             /* CUBIC SPLINE TABLE DISPERSION */
 540             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 541             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 542             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 543             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 544             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 545             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 546             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 547             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 548             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 549
 550             /* CUBIC SPLINE TABLE REPULSION */
 551             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 552             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 553             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 554             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 555             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 556             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 557             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 558             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 559             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 560             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 561
 562             fscal            = _fjsp_add_v2r8(felec,fvdw);
 563
 564             /* Update vectorial force */
 565             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 566             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 567             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 568
 569             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 570
 571             /* Inner loop uses 57 flops */
 572         }
 573
 574         if(jidx<j_index_end)
 575         {
 576
 577             jnrA             = jjnr[jidx];
 578             j_coord_offsetA  = DIM*jnrA;
 579
 580             /* load j atom coordinates */
 581             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 582                                               &jx0,&jy0,&jz0);
 583
 584             /* Calculate displacement vector */
 585             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 586             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 587             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 588
 589             /* Calculate squared distance and things based on it */
 590             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 591
 592             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 593
 594             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 595
 596             /* Load parameters for j particles */
 597             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 598             vdwjidx0A        = 2*vdwtype[jnrA+0];
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 608             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 609                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 610
 611             /* Calculate table index by multiplying r with table scale and truncate to integer */
 612             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 613             itab_tmp         = _fjsp_dtox_v2r8(rt);
 614             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 615             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 616             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 617
 618             vfconv.i[0]     *= 8;
 619             vfconv.i[1]     *= 8;
 620
 621             /* COULOMB ELECTROSTATICS */
 622             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 623             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 624
 625             /* CUBIC SPLINE TABLE DISPERSION */
 626             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 627             F                = _fjsp_setzero_v2r8();
 628             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 629             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 630             H                = _fjsp_setzero_v2r8();
 631             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 632             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 633             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 634             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 635
 636             /* CUBIC SPLINE TABLE REPULSION */
 637             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 638             F                = _fjsp_setzero_v2r8();
 639             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 640             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 641             H                = _fjsp_setzero_v2r8();
 642             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 643             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 644             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 645             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 646             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 647
 648             fscal            = _fjsp_add_v2r8(felec,fvdw);
 649
 650             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 651
 652             /* Update vectorial force */
 653             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 654             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 655             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 656
 657             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 658
 659             /* Inner loop uses 57 flops */
 660         }
 661
 662         /* End of innermost loop */
 663
 664         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 665                                               f+i_coord_offset,fshift+i_shift_offset);
 666
 667         /* Increment number of inner iterations */
 668         inneriter                  += j_index_end - j_index_start;
 669
 670         /* Outer loop uses 7 flops */
 671     }
 672
 673     /* Increment number of outer iterations */
 674     outeriter        += nri;
 675
 676     /* Update outer/inner flops */
 677
 678     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 679 }